基于HSI高光谱数据的水稻光谱特征分析与识别技术研究

【作者】 贾德伟；

【导师】 钟仕全；

【作者基本信息】 广西师范学院 ， 地图学与地理信息系统， 2011， 硕士

【摘要】 水稻是中国的主要粮食作物,实时准确地获取水稻面积等信息,有利于国家粮食安全和贸易安全的保证。传统水稻遥感监测主要集中利用TM、CBERS星的CCD数据以及MODIS数据。HSI数据是国内首个航天高光谱数据,国内对HSI数据在水稻分类识别中的研究几乎空白。以广西玉林市-博白地带2010年10月22日的一景HSI数据为研究区,主要进行了如下研究工作:(1)HSI数据介绍和预处理研究:对HSI数据的特点、命名规则等进行介绍。针对HSI数据产品特点,对HSI数据进行包括数据格式转换、绝对辐射亮度值变换、FLAASH大气校正、质量差波段去除和几何精校正等预处理工作。(2)晚稻等地物类光谱特征分析:数据预处理后,根据研究区概况和同时相HJ-1A星CCD多光谱数据情况,获取晚稻等地物的空间信息,并得到不同地物在HSI数据的光谱变化曲线,分析晚稻与其它地物的光谱差异,为波段选择做好铺垫。(3)波段选择方法研究:通过对信息量和类别可分性波段选择方法进行研究,在此基础上,提出改进的基于信息量和类别可分性的组合波段选择方法,最后选择得分值排序靠前的30个波段作为波段选择结果,为后面的精准分类做了铺垫。(4) SVM在研究区晚稻识别研究:基于SVM和MLC理论,进行HSI数据的晚稻识别,得出波段选择后,SVM对HSI数据总体分类精度最高,其次为波段选择后的MLC方法,而基于非波段选择数据的分类准确率低于波段选择后的结果,证明波段选择在高光谱遥感分类中的必要性,并说明SVM在高光谱分类中的优越性;在SVM理论下,HSI数据晚稻识别精度和同时相CCD数据对比,得出前者总体精度为89.46,略低于后者3.36个百分点,但其仍具有较高的分类精度,可为进一步的定量遥感奠定基础。更多还原

【Abstract】 Rice is the main grain crop in china, obtaining punctually and exactly rice area and other information is beneficial for the guarantee of the country’s food supply security and trade security.The traditional remote-sensing monitoring of rice mainly focused use of TM imagery、CCD imagery of CBERS satellite and MODIS data. The HSI imagery of HJ-1A Satellite is not researched in domestic rice identification. the region of yulin city-bobai country in guangxi was used as research area,many works was done as follows:(1)HSI Imagery Introduction and Proprecessing:the article introduced the character of HSI imagery,etc.some proprecessing work were done,such as data format transformation , absolute radiance value transformation,FLAASH atmospheric correction,poor quality band removal and Geometric Correction,etc.( 2 ) The Spectral Difference of Rice and other geo-objects : After preprocessing of HSI imagery, the distribution of rice and other geo-objects were obtained according to the summary of the research area and the Multispectral CCD Imagery of HJ-1A Satellite,The analysis of spectral difference was researched based on the spectral curve of different geo-objects,which laid a strong theoretical foundation for Band Selection .(3)The research of Band Selection methods:Band Selection methods were introduced based on the Information of Band and the Class Separability. the innovative approach of Band Selection,the combination of Band Informationt and Class Separability,was raised by the foundation of the above methods. The top 30 band of HSI imagery were reserved as the result of band selection according to the score of different bands in some two geo-objects,which made good bedding for the following Classification.(4)The Identification of Rice in research area based on the method of SVM: The Identification of Rice in HSI Imagery were introduced by the methods of SVM and MLC.the collusion was as follows:In classification accuracy,after Band Selection,the result of HSI imagery based on SVM is the best result,then, the result of HSI Imagery based on MLC is the better result,finally, because of not Band Selection,the result of HSI Imagery is lower,but the result of SVM is higher than MLC under the circumstance.So,it is essential to select Band in Hyperspectral Remote Sensing,and the SVM have the advantage of Hyperspectral Imagery Classification. Under the guidance of SVM method,compared the accuracy of Rice in HSI Imagery with the accuracy of Rice in the same temporal CCD Imagery,the overall accuracy of the former which was 89.46% was lower than the latter,but the HSI imagery still had the good classification accuracy.更多还原